Questions: Variant Calling and Genome-Wide Association Studies

Question 1

Why does GATK's variant calling pipeline use a Bayesian framework rather than simply counting how many reads show the reference versus alternate allele?

• A. Bayesian methods are faster computationally
• B. The Bayesian framework integrates base quality scores, mapping quality, and prior expectations to distinguish true variants from sequencing errors
• C. Simple allele counting cannot detect indels
• D. Bayesian methods do not require a reference genome

Answer: (B)

Raw allele counts are confounded by sequencing errors, mapping artifacts, and variable coverage. A Bayesian model incorporates the probability that each base call is correct (base quality), the confidence of the read's genomic placement (mapping quality), and prior expectations about variant frequency. It outputs a posterior probability for each possible genotype (homozygous reference, heterozygous, homozygous alternate), providing a principled way to call variants even at modest coverage where a simple majority-vote approach would be unreliable.

Question 2

A GWAS identifies a SNP significantly associated with type 2 diabetes. This SNP is the mutation that causes the disease.

• T. True
• F. False

Answer: False

The associated SNP is typically a tag SNP — one of many correlated variants in a region of linkage disequilibrium. The causal variant may be the tag SNP itself, but more often it is a different variant in the same LD block. Furthermore, most GWAS hits fall in noncoding regions and likely affect gene regulation rather than protein sequence. Fine-mapping, functional studies, and eQTL (expression quantitative trait locus) analyses are needed to move from a GWAS association signal to the actual causal variant and mechanism.

Question 3

Explain why the genome-wide significance threshold for GWAS is typically set at p < 5e-8 rather than the conventional p < 0.05.

Model answer: A GWAS typically tests approximately 1 million independent SNPs (after accounting for linkage disequilibrium) for association with a phenotype. At p < 0.05, this would produce ~50,000 false positives. The threshold of 5e-8 approximates a Bonferroni correction for 1 million tests (0.05 / 1,000,000 = 5e-8), controlling the genome-wide false positive rate at 5%. This stringent threshold means that only very strong statistical signals survive, requiring large sample sizes (often tens to hundreds of thousands of individuals) to detect the typically small effects of individual variants on complex traits.

The 5e-8 threshold has become a community standard for human GWAS. It is conservative (Bonferroni assumes independence, but SNPs in LD are correlated), but its stringency has proven valuable — GWAS results that pass this threshold replicate in independent cohorts at very high rates, demonstrating that the threshold effectively controls false discoveries.